C300 Sensor

[Jon Fairhurst]

Here are some great articles describing the C300 sensor. Clearly, this sensor will not be limited to the C300.

A nice overview...
Image Sensors World: Canon Super-35mm Cinema Sensor Explained

The sensor, in detail...
http://learn.usa.canon.com/app/pdfs/..._Sensor_WP.pdf

The back end...
http://learn.usa.canon.com/app/pdfs/...ristics_WP.pdf

[Ken Diewert]

Jon,

There's a reason that some of us rely on opinion leaders to read and decipher literature like that, before maybe camera purchase decisions. Thanks for posting. I'll let the pixel peepers have at it, and I'll just look at the pretty pictures for awhile : )

[David Heath]

To start off with, I expect the performance from this sensor to be probably as good as can be hoped for for the 1080 system, but it would be foolish to expect it to be "perfect". (Whatever that means.) It will be interesting to see proper tests, and in the context of those articles, the most interesting tests may be zone plates, and what aliasing is visible. (Because there will inevitably be some, even if pretty irrelevant for real-world usage.)

In particular, it will be how it handles very fine detail, finer than the 1080 system can resolve, and if anybody has plans to test it with a zone plate I hope they use one with detail up to 2160 lpph. I expect it to deal with the range up to 1080 well - it's the range 1080-2160 that will be interesting. Why? Imagine 2160 horizontal lines, 1080 white, 1080 black, and think of a situation where they exactly line up with the sensor rows. With the way the sensor reads out, that would theoretically give either only red and green photosites being illuminated or blue and green - depending which rows are white, which black. So theoretically either a uniform yellow or cyan output!!

Practically, it's not likely to happen exactly as that because of a low pass filter of course. But that raises another interesting question. I'd previously thought that a future camera may use the same sensor, and offer the choice of reading out as the C300 for 1080 mode, or fully deBayering for a 4k mode. It now occurs to me that that may be difficult. The optical low pass filter for each case would need to be very different - to allow sufficient detail through for 4k would likely lead to excessive aliasing in 1080 mode. Or am I overlooking something?

[Jon Fairhurst]

Good point, David. The anti aliasing filter might indeed need to be changed for an optimum 4K implementation. For the 1080 implementation, we need a fairly soft filter to keep the red and blue from aliasing. For a de-bayer implementation, we want some per-channel aliasing since the idea is to imply information about a given channel from the other two channels.

On the other hand, optical low pass filters are pretty basic. As I understand it, they use linear polarizers to split the light and they shift one of the images. It ends up being like a two tap digital filter, aided by a bit of analog smear. They aren't exactly brick wall filters. There's a significant transition band. So, to make a sharp 1080p signal, there will be a fair amount of aliasing. By the same token, if you want to fully eliminate aliasing, the image will be soft.

My guess is that the filter is tuned for a sharp 1080p output with some level of acceptable aliasing. Pulling 4K will have virtually no aliasing, but would be a bit softer than true 4K. Because of the gradual transition band, Canon can split the difference, rather than have to choose 1080 or 4K.

However, I could be wrong. Canon might indeed make two versions of the chip with different OLPF layers. I doubt it though. If users have been tolerant enough to shoot features and episodic TV on DSLRs, we can certainly accept the C300 with the filter tuned slightly toward the sharp side.

Like you, I'd love to see some zone plates. :)

[Sareesh Sudhakaran]

Cool links. Thanks for sharing, Jon.

[Roger Keay]

The Canon document titled 'NEW 35MM CMOS IMAGE SENSOR FOR DIGITAL CINE MOTION IMAGING' states on page 7 that "The design strategy of this new sensor is to not to seek any form of “4K” resolution — but rather to specifically confine the reconstruction of each of the R, G, and B video components to a full digital sampling structure of 1920 (H) x 1080 (V) — according to the SMPTE 274M HDTV Production Standard." The logical consequence of this decision is an OLPF to minimize aliasing for standard HDTV. The double active H and V pixel dimensions of the sensor with the dual green configuration oversamples the image and delivers higher luminance resolution within the 1080 HD format than a conventional 3 chip camera.

Canon's decision to develop a sensor for a Cine camera tailored for 1080 HDTV makes sense when the global volume of cinema style production for television is considered. The underlying photosite array has potential for a Bayer process '4K' cine camera possibly using different readout sequence, CFA filters (wider colour gamut) and OLPF which would make it a different sensor than the component in the C300. A 4K sensor using the same non-Bayer design concepts as the C300 sensor would require over 8 thousand photosites horizontally and 4 thousand vertically for a total of over 32 million photosites.

[Jon Fairhurst]

Good catch, Roger!

[Alister Chapman]

I think Canon's key reason for this sensors pixel count and design is to allow them to use their existing video processor chain from the XF300 camcorders, making it quicker and cheaper for them to bring out this new camera. I'm not saying that this necessarily makes it a bad or compromise design, just that that's why I think they did it this way.

[David Heath]

Originally Posted by Roger Keay

The Canon document ...... states on page 7 that "The design strategy of this new sensor is to not to seek any form of “4K” resolution — but rather to specifically confine the reconstruction of each of the R, G, and B video components to a full digital sampling structure of 1920 (H) x 1080 (V) ..........

Originally Posted by Alister Chapman

I think Canon's key reason for this sensors pixel count and design is to allow them to use their existing video processor chain from the XF300 camcorders, making it quicker and cheaper for them to bring out this new camera......

I think it's worth reminding ourselves of exactly what this sensor is in terms of dimensions - 4206 (H) x 2340 (V) - *NOT* 3840x2160, although the latter figure is indeed the area that is read in the case of the C300. It's normal for the total photosite count to exceed the active count - but not by as much as this. Consequently, I would put a pretty large bet that the basic construct of the sensor is PRIMARILY with an eye to the future and 4k, and conventional deBayering. If it was true that "the design strategy of this new sensor is to not to seek any form of “4K” resolution....." it would make more sense to have photosite dimensions more like 3880x2190 to minimise the "windowing" effect and make optimum use of the full s35 area.

In the meantime, direct read of the 3840x2160 window allows the re-use of existing technology (video processor, codec etc) to bring a product to market relatively quickly and cheaply - I think the three of us are in agreement that this is highly sensible. I fully agree that the main market at the moment is decent quality 1080 for broadcast standard acquisition - and all the indications are that the C300 is fully capable of giving just that, and a broadcast codec inbuilt as well.

Like Alister, then neither do I mean any of what I say critically, quite the opposite really. The only element of compromise may seem to be the the small amount of cropping inherent in the windowing - but that's a very small price to pay for the potential futureproofing of being able to use the same basic chip in a future 4k camera.

Compared to the F3, I doubt zone plates will show much difference for luminance resolution, but I'll be interested to see the difference in aliasing. My bet is that the approx 8 versus 3 megapixel counts will make the C300 better in this respect. We'll see.